← Back
OpenAI
OpenAI releases GPT-5.4 mini and nano; smaller models match larger variant performance on coding benchmarks
OpenAI APIChatGPTOpenAI · releasemodelapifeaturepricing · openai.com ↗

Overview

OpenAI today released GPT-5.4 mini and GPT-5.4 nano, two smaller models optimized for high-volume, latency-critical workloads. These models bring significant improvements over their GPT-5 predecessors while maintaining the speed and efficiency necessary for responsive coding assistants, subagents, and real-time multimodal applications.

Model Capabilities

GPT-5.4 mini delivers substantial improvements across multiple dimensions:

  • Coding Performance: Achieves 54.4% on SWE-Bench Pro (vs. 45.7% for GPT-5 mini) and 60% on Terminal-Bench 2.0, approaching the full GPT-5.4 model's 57.7% while running 2x faster
  • Reasoning & Intelligence: Scores 88% on GPQA Diamond and strong performance on tool-calling benchmarks (57.7% on MCP Atlas)
  • Computer Use: 72.1% on OSWorld-Verified, substantially outperforming GPT-5 mini's 42%
  • Multimodal Understanding: Strong vision performance with 78% on MMMUPro with Python

GPT-5.4 nano targets cost-sensitive tasks like classification, data extraction, and simple subagent workflows, offering further improvements over GPT-5 nano while being the cheapest option available.

Availability & Pricing

Model Input Cost Output Cost Availability
GPT-5.4 mini $0.75/1M tokens $4.50/1M tokens API, ChatGPT, Codex
GPT-5.4 nano $0.20/1M tokens $1.25/1M tokens API only

Both models support the full suite of OpenAI capabilities: tool use, function calling, web search, file search, computer use, and a 400k context window (mini). In Codex, GPT-5.4 mini uses only 30% of the GPT-5.4 quota.

Key Use Cases

Coding Workflows: Fast iteration on targeted edits, codebase navigation, front-end generation, and debugging loops benefit most from these models' low-latency responses.

Subagent Architectures: Developers can now compose hybrid systems where GPT-5.4 handles planning and coordination while delegating narrower subtasks to GPT-5.4 mini subagents running in parallel—improving throughput and reducing costs.

Computer Use: The mini model excels at interpreting dense UI screenshots and completing computer use tasks with minimal latency overhead.

Immediate Action Items

  • API Users: GPT-5.4 mini and nano are available immediately via the OpenAI API
  • ChatGPT Users: GPT-5.4 mini is available in the "Thinking" feature for Free and Go subscribers
  • Codex Users: Deploy GPT-5.4 mini across the Codex app, CLI, and IDE extensions at reduced quota consumption
  • Review Security: Check the System Card addendum on the Deployment Safety Hub for safeguard details